Roumell Asset Management commentary for the second quarter ended June 30, 2016.
Roumell Asset Management – Forecasting and Deep Value Investing
In the recently published book, Superforecasting, The Art and Science of Prediction, authors Philip E. Tetlock and Dan Gardner report on the rich details of their Good Judgement Project (GJP). The GJP was part of a larger longitudinal study involving thousands of individuals over several years to better understand forecasting with the ultimate goal of increasing the U.S. intelligence community’s forecasting accuracy. The book’s authors are experts at analyzing forecasting abilities and offer their views on whether anyone can forecast meaningfully above average (yes, but it’s a small group) and describe the attributes that they believe help explain these superforecasters’ abilities. Tetlock and Gardner believe that the habits of superforecasters can be codified and taught and end their book with a Ten Commandments of superforecasting for their readers.
RAM has always been forecasting-averse, and for good reason—it’s very difficult and the odds of success are low. In Tetlock and Gartner’s view, about 2% of their participants (a group comprised of engineers, lawyers, artists, scientists, Wall Streeters and Main Streeters, professors and students), qualified as superforecasters. These were well-read, smart people who stay on top of world affairs and know how to research questions. This rather sobering notion is precisely why RAM has long de-emphasized highly liquid markets/securities, growth investments highly dependent on predicting future earnings, forecasting future commodity prices or the direction of interest rates.
To be clear, all investments at some level rely on certain forecasts. RAM’s goal has always been to divide investment narratives into essentially two buckets—what is known today and what may materialize tomorrow. The more an investment thesis rests on the former, “what is,” the more we like it. In our last quarterly letter we discussed, in some detail, buying double-discounted closed-end bond funds possessing a discount to NAV and an NAV itself that is reflective of a portfolio of bonds trading at a deep discount to par value wherein we effectively were able to purchase a diversified portfolio of high yield bonds at seventy cents of par value. These investments are classic “what is” investments. The investment relies far more on a discount to value today as opposed to possible value creation tomorrow that is heavily dependent on forecasting.
Nonetheless, even our double discounted closed-end bond fund investments have some degree of forecasting built into the underlying investment thesis. For instance, we modeled portfolio default rates of 5% to 20% (with zero recovery values), which means a tsunami of defaults exceeding our high-end model would be problematic; a low probability, but something north of zero. Moreover, we predict the portfolio managers will not trade the portfolio in a way that eliminates the value of the double-discount, i.e., terrible bond trading that renders moot an analysis of the existing portfolio. Thus, while all investments do involve forecasting, our approach has been to minimize the requirement to forecast the future as much as possible.
The Intelligence Advanced Research Projects Activity (IARPA) is an agency within the Intelligence Community (IC) that reports to the director of National Intelligence. Its job is to increase the accuracy of American intelligence estimates. Not surprisingly, no one really knows how good the overall intelligence forecasting is because it’s never been measured, likely the result of analysts not wanting the light shined on their significant, but often not useful, efforts. The IC was humiliated by its conviction that Iraq possessed WMD, adding to other big “misses” such as the surprise collapse of the Soviet Union. With a desire to better understand the business of predicting the future, IARPA created a forecasting tournament comprised of five teams led by top researchers to measure forecasting acumen among intelligence professionals and educated, informed, common citizens.
The GJP was one team comprised of 2,800 individuals from varying backgrounds who were selected by the authors. Leveraging their knowledge and prior research on the subject of forecasting, the authors put in place a structure for the GJP team. From September 2011 to June 2015, teams were required to submit daily forecasts for nearly 500 questions about world affairs. Participants were allowed to regularly update and change their forecasts; each change becoming a new forecast. Questions like the following were posited: Will Greece leave the Eurozone?; Will Israel attack Iranian nuclear facilities by September?; and, Will Saudi Arabia cut their oil production output by the end of this year? The study rewarded confidence/conviction levels such that an individual assigning an 80% probability to a potential event occurring received a higher score than someone assigning a 50% probability to the same event if it, in fact, occurred.
The GJP group beat the official control group (comprised of an IC team operating under the same constraints) by 60% after year 1 and by 78% by the end of year 2. The GJP team also beat universityaffiliated teams, including the University of Michigan and MIT, from 30% to 70%, and outperformed professional intelligence analysts with access to classified information. The study’s basic conclusion: generating above average forecasting value is unlikely, but possible. To wit, roughly 2% of the individuals in the study showed themselves to be superforecasters. What were some of the traits and habits shared among this group of superforecasters?
One of the big takeaways from the authors’ history of studying forecasting is comparing one group they describe as hedgehogs to another who they describe as foxes. Superforecasting, their research shows, is highly correlated with how one thinks, not what one thinks. Hedgehogs tend to think around “big ideas” and present as highly confident people; to their forecasting detriment. Referring to this group, the authors state, “They sought to squeeze complex problems into the preferred cause-effect templates and treated what did not fit as irrelevant distractions…they were unusually confident and likelier to declare things ‘impossible’ or ‘certain.’ Committed to their conclusions, they were reluctant to change their minds even when their predictions clearly failed. They would tell us, ‘Just wait.’”
The other group, foxes, was comprised of much more pragmatic thinkers, who were often humble in their assessments. Referring to this group, the authors noted, “These experts gathered as much information from as many sources as they could. They talked about possibilities and probabilities, not certainties. And while no one likes to say ‘I was wrong,’ these experts more readily admitted it and changed their minds.”
While reading Superforecasters one is reminded of the difficulty of forecasting, particularly the macroeconomic variety, and why RAM has tried to stay clear of being overly dependent on such forecasting as much as possible. We are in the business of forecasting at some level, but the goal has always been to keep it to a minimum while pursuing existing embedded value. That means not owning investments wholly dependent on rising/falling commodity prices, interest rates predictions, estimating GDP growth and the like.
On an individual security basis, it means minimizing our dependence on growth projections and preferring to emphasize “what is”, while hopefully owning optionality without paying for it. It’s why we owned a gold streaming company (on two separate occasions) when it provided a 12% plus free cash flow yield, but sold it when